AI agents turn Kokonut farm data into faster coordination.

Farms produce information constantly: satellite images, field logs, soil readings, drone surveys, crop cycles, harvest records, MRV payloads, and DAO reporting requirements. Today, much of that work still depends on manual handoffs. An operator collects data, someone processes it, someone else reviews it, another person formats it for a report, and the result may only become useful after long delays. Kokonut’s agent layer is designed to reduce those handoffs. Agents can ingest farm data, calculate metrics, prepare structured records, route outputs for review, trigger payments, and help convert farm activity into evidence that DAO members, Guilds, farm operators, and public goods funders can actually use.

See what agents can do

Explore MRV automation, harvest forecasting, impact scoring, grant drafting, proposal drafting, and farm monitoring.

Build your first agent

Follow the safe builder path: choose a task, define inputs and outputs, connect to Kokonut data, test, and request review.

The Agentic Marketplace is still in development. Treat contract addresses, service registry flows, marketplace listings, escrow flows, and autonomous payments as developing infrastructure until testnet or mainnet deployments are published. Do not describe planned agent capabilities as live farm guarantees.

What this page is for

Reader	What you should get from this page	Best next step
Agent builder	Understand safe first agent opportunities and integration requirements	Start with the MRV Reporter or Harvest Forecaster path
Data / MRV engineer	Understand how agents interact with Farm Registry records, IPFS payloads, EAS attestations, and the Kokonut Hub	Review the MRV workflow
Farm operator	Understand which farm tasks agents can help automate	Identify repeatable tasks worth automating
DAO or Guild reviewer	Understand what needs human review before agent outputs become trusted	Use the review checklist
Partner or funder	Understand how automation supports evidence, not hype	Compare estimates, measurements, attestations, and reports

Agents are not a replacement for agronomists, farm operators, DAO governance, or human review. They are coordination tools that help standardize repeatable work and move evidence through the Kokonut system faster.

AI agents at a glance

Layer	What it does	Status
Farm data	Reads farm records, crop cycles, harvest events, MRV payloads, and public Data Hub entries	Live data exists for Adelphi; integrations expanding
Intelligence layer	Structures, operational, environmental, financial, governance, and Web3 records for analysis	Active infrastructure
Agent runtime	Runs task-specific services such as MRV reporters, harvest forecasters, and impact scorers	Developing
Identity	Registers agents with names, operators, capabilities, and service metadata	Developing
Payments	Uses USDC and x402-style flows for paid service calls	Developing
Verification	Connects outputs to IPFS/Filecoin records, Farm Registry events, EAS attestations, and reviewer approval	MRV pipeline active; agent automation developing
Reputation	Tracks completed work, quality, disputes, and longer-term contribution history	Developing

The core idea

The goal is not to let agents invent impact claims. The goal is to help agents move real farm events through a structured evidence pathway.

What agents should automate first?

Agents should begin with repeatable, bounded tasks where the input, output, review rule, and evidence requirement are clear.

MRV Reporter

Ingests satellite, drone, soil, or field observations and prepares structured MRV payloads for review and attestation.

Harvest Forecaster

Applies registered crop assumptions to estimate expected harvests, loss rates, revenue scenarios, and actual-vs-forecast differences.

Impact Scorer

Aggregates verified farm records across a reporting period and prepares draft EBF, SDG, or public-goods reporting summaries.

Grant Drafter

Pulls structured farm data, MRV records, SDG alignment, and proof points into draft grant applications for human editing.

Proposal Drafter

Converts farm records, budgets, milestones, and evidence into DAO proposal drafts for sponsor review.

Cross-Farm Monitor

Compares vegetation, harvest, soil, and operations data across farms to flag anomalies, missing records, or possible intervention needs.

Live vs. developing

Area	How to describe it
Adelphi farm data	Live reference project with public records and MRV workflows
MRV methodology	Active trust pipeline for turning farm activity into structured evidence
Common Data Schema	Live Framework primitive for standard farm records
Kokonut Intelligence	Active data backbone for farm, governance, environmental, financial, and Web3 activity
Agentic Marketplace	In development, use the development branch and avoid claiming deployed marketplace behavior until addresses are published
x402 payments	Design direction for autonomous service payment; implementation should be tested before production claims
Agent reputation	Developing should distinguish task history, review outcomes, Guild recognition, and DAO-approved rewards
Carbon or credit-related automation	Research / methodology-dependent; agents can support evidence collection, but cannot create certified climate claims by themselves

A useful agent can produce a draft, a calculation, an alert, or a structured payload. A trustworthy Kokonut claim still needs source data, review, evidence storage, attestation, and public reporting.

Agent system architecture

The four builder primitives

Primitive	Plain English	In Kokonut
Identity	Who is the agent and who operates it?	Agent name, operator wallet, capability manifest, status, and service metadata
Capability	What can the agent do?	Inputs, outputs, pricing, SLA, review rules, and artifact format
Payment	How does the agent get paid?	USDC / x402-style calls, escrow for longer tasks, and completion proof requirements
Evidence	Why should anyone trust the output?	Source data, IPFS/Filecoin CID, Farm Registry event, EAS attestation, reviewer approval, and public Data Hub record

Capability manifest

Every agent should declare what it accepts, what it returns, and what review evidence is required.

{
  "agent_name": "kokonut-mrv-reporter",
  "version": "0.1.0",
  "status": "development",
  "description": "Prepares structured MRV payloads from remote sensing or field data for review and attestation.",
  "inputs": {
    "farm_id": {
      "type": "string",
      "required": true,
      "description": "Farm Registry identifier, such as adelphi"
    },
    "period_start": {
      "type": "string",
      "format": "ISO8601",
      "required": true
    },
    "period_end": {
      "type": "string",
      "format": "ISO8601",
      "required": true
    },
    "imagery_source": {
      "type": "string",
      "enum": ["sentinel", "landsat_8", "drone", "field_upload"],
      "required": true
    }
  },
  "outputs": {
    "mrv_event_draft_id": {
      "type": "string",
      "description": "Draft event created before final review"
    },
    "vegetation_indices": {
      "type": "object",
      "properties": {
        "ndvi": { "type": "number" },
        "ndre": { "type": "number" },
        "reci": { "type": "number" },
        "msavi": { "type": "number" }
      }
    },
    "artifact_cid": {
      "type": "string",
      "description": "IPFS/Filecoin CID for the source artifact or generated report"
    },
    "review_status": {
      "type": "string",
      "enum": ["draft", "needs_review", "approved", "rejected"]
    }
  },
  "review_required": true,
  "reviewer_role": "Impact Guild or authorized MRV reviewer"
}

Keep manifests versioned. If an agent changes input fields, output fields, pricing, or review rules, publish a new version and document the migration impact.

MRV automation workflow

Agents should compress manual work without bypassing verification.

Ingest source data

Pull satellite imagery, drone uploads, soil readings, harvest logs, or field observations. Record the source, timestamp, farm identifier, and collection method.

Calculate structured metrics

Calculate vegetation indices, crop estimates, loss-rate scenarios, soil trends, or reporting summaries using documented formulas and farm boundaries.

Create a draft payload

Format results into a Farm Registry-compatible draft. Mark it as a draft until a reviewer approves it.

Attach evidence

Store source files or generated reports with IPFS/Filecoin CIDs so reviewers can inspect the evidence behind the output.

Request review

Route the draft to the relevant Guild, farm operator, or authorized reviewer. Agents should not self-certify high-impact claims.

Publish and attest

After approval, submit the final event to the Farm Registry and create the EAS attestation that links the public claim to its evidence.

Vegetation indices agents may compute

Index	Use case	Review note
NDVI	General vegetation health across a farm period	Useful, but should be interpreted alongside crop stage and ground context
NDRE	Canopy maturity and chlorophyll stress in later growth stages	Works best when crop canopy is developed
ReCI	Chlorophyll and nitrogen-related stress signal	Requires agronomic context before recommendations
MSAVI	Early growth, sparse canopy, and exposed soil conditions	Better for early-stage plots than NDVI alone

Vegetation indices are signals, not final conclusions. Agents should flag potential conditions and route them for review by a farm operator or agronomist before making operational recommendations.

Payments and escrow

Payment automation should match task complexity.

Task type	Suggested payment pattern	Why
Fast API call	x402-style pay-per-request	Good for small forecasts, metadata lookups, or simple calculations
Long-running analysis	Escrow with completion proof	Better for drone processing, multi-farm reports, or large grant drafts
Recurring monitoring	Milestone or subscription-style budget	Useful for weekly or monthly farm monitoring tasks
High-risk claims	Payment after review	Avoids rewarding outputs that fail evidence or quality checks

Keep payment flows separate from evidence quality. A paid output is not automatically a verified output.

Reputation and quality control

Agent reputation should be based on useful work, not just task volume.

Signal	What it means	Who uses it
Task completion	The agent returned an output before the deadline	Builders, marketplaces, reviewers
Review outcome	A human or Guild reviewer accepted, requested changes, or rejected the output	Guilds, DAO members, operators
Evidence quality	Source data, formulas, CIDs, and assumptions were inspectable	MRV reviewers, grant reviewers, researchers
Dispute history	Outputs were challenged or escalated	Marketplace users and DAO reviewers
Long-term contribution	The agent’s work consistently supported farm reporting or operations over time	Guilds and DAO proposal reviewers

Agents that consistently produce useful, reviewable, and evidence-backed outputs can support Guild contribution records. Larger recognition or rewards should still follow the relevant Guild or DAO proposal process.

Builder starts path

Pick one bounded task

Start with one repeatable job, such as preparing an MRV draft, calculating a harvest forecast, or formatting a proposal draft.

Define the input and output schema

Write the capability manifest before writing complex code. Reviewers should know exactly what the agent accepts and returns.

Connect to one Kokonut data source

Use the Common Data Schema, Farm Registry API, Kokonut Hub, or Kokonut Intelligence as the source of truth. Do not scrape or reinterpret records without documenting the source.

Mark outputs as a draft by default

Agent outputs should be treated as drafts until a reviewer approves them, especially for MRV, impact, finance, carbon, governance, or grant claims.

Add evidence artifacts

Store source files, generated reports, formulas, and assumptions with CIDs or stable references so outputs are reviewable.

Open an issue or Guild review request

For non-trivial integrations, open an issue before a pull request. Schema, MRV, attestation, and payment changes need compatibility review.

Safe first agent ideas

Agent	Input	Output	Reviewer
MRV Draft Reporter	Farm ID, period, imagery, or field record	Draft MRV payload and evidence CID	Impact Guild / MRV reviewer
Harvest Actuals Checker	Crop cycle, planted area, harvest log	Actual-vs-forecast report	Farm operator / Finance Guild
Proposal Formatter	Farm record, budget, milestones, evidence	Draft proposal in Kokonut template format	Governance Guild
Data Quality Monitor	Farm Registry records and Data Hub entries	Missing fields, stale data, or inconsistency alerts	Technology Guild
SDG Evidence Assistant	MRV records, harvests, jobs, and public goods allocation	Draft SDG evidence map	Impact Guild
Grant Proof Assistant	Farm summary, MRV records, SDG map, photos	Draft proof package for grant submission	Communications Guild

What agents should not do

Do not let agents…	Safer alternative
Certify carbon credits	Collect evidence that could support future methodology review
Claim guaranteed yields or revenue	Compare forecasts against actual records and show assumptions
Publish impact claims without evidence	Route drafts through MRV, reviewer approval, and EAS attestation
Move DAO funds autonomously	Draft proposals for human sponsorship and DAO vote
Change the Common Data Schema silently	Open a Framework Upgrade Proposal with a migration plan
Treat planned contracts as deployed	Label infrastructure as live, developing, planned, or deprecated
Replace farm operators	Assist operators with alerts, summaries, and structured reports

How this connects to Kokonut Intelligence

Kokonut Intelligence is the data backbone that can make agents useful. It gives agents access to structured context instead of forcing them to infer from scattered documents.

Intelligence function	Agent benefit
Farm operations records	Agents can reference actual activity instead of inventing assumptions
Environmental measurements	Agents can calculate trends and flag anomalies
Financial and harvest records	Agents can compare forecasts against actuals
Web3 and DAO records	Agents can draft proposals and reporting summaries with governance context
Queryable analytics layer	Agents can answer structured questions for builders, reviewers, and operators

Read Kokonut Intelligence →

Review the checklist before using an agent output

Does the output cite or reference the source farm record?
Does it distinguish between forecast, estimate, measurement, attestation, and verified report?
Does it include timestamps, farm ID, source type, and reviewer status?
Is the formula or model documented?
Are source artifacts linked through stable storage or CIDs?
Has a human or Guild reviewer approved high-impact claims?
Does the output avoid unsupported financial, carbon, health, or climate claims?
Does it avoid describing development infrastructure as live?

Developer resources

Resource	Purpose
Build with Kokonut	Repositories, deployed contracts, data primitives, and contribution workflow
Farm Registry API	API reference for farm records, MRV events, harvests, attestations, and impact endpoints
MRV methodology	Evidence pipeline for farm activity, payloads, IPFS records, EAS attestations, and reports
Common Data Schema	Baseline farm record consumed by DAO members, developers, agents, and dashboards
Kokonut Intelligence	Canonical data backbone for farm, environmental, financial, governance, and Web3 activity
Kokonut Guilds DAO	Contribution path for builders who want Guild recognition and possible DAO rewards
Glossary	Definitions for MRV, EAS, x402, ERC-8004, Guilds, and related terms

Build with Kokonut

Start with repos, contribution workflow, data primitives, and live vs. developing infrastructure.

Read the MRV Methodology

Understand how agent outputs should become structured evidence instead of unsupported claims.

Open the Agentic Marketplace repo

Review the marketplace codebase, contracts, frontend, and OpenServ integration work.

Join the Technology Guild

Contribute useful agent tooling, earn standing through work, and route larger contributions through DAO governance.

Playbooks

Kokonut × AI Agents

AI agents turn Kokonut farm data into faster coordination.

See what agents can do

Build your first agent

What this page is for

AI agents at a glance

The core idea

What agents should automate first?

MRV Reporter

Harvest Forecaster

Impact Scorer

Grant Drafter

Proposal Drafter

Cross-Farm Monitor

Live vs. developing

Agent system architecture

The four builder primitives

Capability manifest

MRV automation workflow

Vegetation indices agents may compute

Payments and escrow

Reputation and quality control

Builder starts path

Safe first agent ideas

What agents should not do

How this connects to Kokonut Intelligence

Review the checklist before using an agent output

Developer resources

Build with Kokonut

Read the MRV Methodology

Open the Agentic Marketplace repo

Join the Technology Guild

​AI agents turn Kokonut farm data into faster coordination.

See what agents can do

Build your first agent

​What this page is for

​AI agents at a glance

​The core idea

​What agents should automate first?

MRV Reporter

Harvest Forecaster

Impact Scorer

Grant Drafter

Proposal Drafter

Cross-Farm Monitor

​Live vs. developing

​Agent system architecture

​The four builder primitives

​Capability manifest

​MRV automation workflow

​Vegetation indices agents may compute

​Payments and escrow

​Reputation and quality control

​Builder starts path

​Safe first agent ideas

​What agents should not do

​How this connects to Kokonut Intelligence

​Review the checklist before using an agent output

​Developer resources

Build with Kokonut

Read the MRV Methodology

Open the Agentic Marketplace repo

Join the Technology Guild

AI agents turn Kokonut farm data into faster coordination.

What this page is for

AI agents at a glance

The core idea

What agents should automate first?

Live vs. developing

Agent system architecture

The four builder primitives

Capability manifest

MRV automation workflow

Vegetation indices agents may compute

Payments and escrow

Reputation and quality control

Builder starts path

Safe first agent ideas

What agents should not do

How this connects to Kokonut Intelligence

Review the checklist before using an agent output

Developer resources